Authors accused of using AI are turning to "paper trail" apps that record how they write
After a publisher pulled a novel over AI claims, tools like Workings and Blindproof log the writing process to prove human authorship. The Authors Guild calls the burden insulting.
DevIsDead Newsroom
Writers spent years defending their books from AI. Now some have to defend themselves against accusations of using it. Miles Klee looks at the new tools meant to help for WIRED.
The trigger
In March, after social media speculation and a New York Times article, Hachette canceled the US release of the horror novel Shy Girl and pulled it from sale in Britain, where it had already been published. The AI detector Pangram estimated about 78% of the text was AI-generated. Author Mia Ballard denies using AI, says an editor made AI-assisted changes to an earlier self-published version, and says she's pursuing legal action. Since then, other authors have faced similar suspicions, including one after winning a short story prize; they deny it too.
Novelist Andrea Bartz, lead plaintiff in the class action that ended in Anthropic's $1.5 billion copyright settlement, says the damage is done once accusations start, and there's no agreed standard for proving human authorship.
The tools
A wave of platforms, among them ObraVera, Human Ink, Wintertext, Blindproof and Workings, track and document how a work gets written, some by encrypting and fingerprinting the text or even typing rhythms. The goal is a record writers can point to if challenged, and proof of provenance when text gets scraped into training data without attribution.
The Authors Guild doesn't endorse any one tool. CEO Mary Rasenberger says it's "insulting" and a burden for authors to have to keep receipts, and that the guild has heard of plenty of human-written work being flagged as AI by Pangram, which claims a 0.01% false-positive rate on creative writing. She urges writers not to accuse each other.
The limits
Process tracking isn't airtight either. Pangram CEO Max Spero, whose company also sells a Google Docs tracking tool, points out that a determined cheater could simply retype AI output by hand, or feed it through an auto-typing tool.
It's a familiar shape: detectors that can't be trusted, followed by surveillance tools to prove innocence. For anyone building creative tools, provenance is becoming a feature users may need, whether or not they want it.
Scoreboard update: Draw. Writers can now prove they're human, at the cost of being watched while they write.
Stay undead.
Comments 0